The invention discloses a prediction method for reservoir thickness which is based on dipole wavelet. Steps are as follows, parameters are extracted, time difference of sound wave between a reservoir and a non-essential reservoir as well as the density are resolved; the time difference of sound wave and the density are converted into a time domain by time-depth conversion, the reflectances of the top and bottom of the reservoir are computed, and a time domain model at the well point is set up, the velocity of longitudinal wave, the density, the wave impedance, and the reflectances are involved in the time domain model; the seismic wavelet is extracted and optimized; a seismic responsing pattern store of reservoir is built by the following steps, the seismic wavelet is taken as a basic wavelet function, the reflectances of the top and bottom of the reservoir are respectively coefficients c and d of the dipole wavelet function to be inputs of the dipole wavelet function, and a reservoir seismic responsing pattern store with a given reservoir section of the work area is set up, spectrum of reservoir thickness is set up; the distribution of the reservoir in three dimensions is predicated quantitatively.